rct-bias-assessment-rob2
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted research papers which could potentially contain hidden instructions designed to influence agent behavior.\n
- Ingestion points: Data is ingested through
scripts/extract_pdf.pyfor PDF files and via trial text provided as input inSKILL.md.\n - Boundary markers: There are no explicit delimiters or safety instructions provided to the agent to distinguish the research content from its operational instructions.\n
- Capability inventory: The skill possesses capabilities to execute shell commands (
python extract_pdf.py) and write to the local file system (saving extracted content to.txtfiles).\n - Sanitization: No sanitization is applied to the extracted text before it is presented to the agent for ROB2 domain assessment.
Audit Metadata